The Engineering Pipeline — From Idea to Production in 24 Hours

Issue #847: AI-powered job priority prediction
- Predict optimal priority level based on prompt complexity
- Endpoint: POST /api/jobs/predict-priority
- Acceptance: accuracy >80% on test set, latency <200ms

# Thom creates the repo on GitLab
glab repo create pyratzlabs/priority-predictor \
  --description "AI job priority prediction service" \
  --visibility private

# Initialize project structure
mkdir -p src tests docs .gitlab
cat > .gitlab-ci.yml << 'EOF'
stages:
  - lint
  - test
  - preview
  - deploy

lint:
  image: python:3.12-slim
  script:
    - pip install ruff black mypy
    - ruff check src/
    - black --check src/
    - mypy src/

test:
  image: python:3.12-slim
  script:
    - pip install -r requirements.txt -r requirements-test.txt
    - pytest tests/ -v --cov=src --cov-report=term-missing
  coverage: '/TOTAL\s+\d+\s+\d+\s+(\d+%)/'

preview:
  stage: preview
  script:
    - vercel deploy --token $VERCEL_TOKEN --yes
  only:
    - merge_requests

deploy:
  stage: deploy
  script:
    - render deploy --service priority-predictor
  only:
    - main
EOF

10:00 — Agent creates Django app skeleton with models, serializers, views
10:45 — ML pipeline: feature extraction from prompts (token count, complexity score, entity detection)
11:30 — Training script: logistic regression on historical job data (2.3M records)
12:15 — API endpoint: /api/jobs/predict-priority with request validation
13:00 — Integration: hook into existing job creation flow
14:00 — Performance optimization: model inference cached, batch prediction support
15:00 — Error handling: graceful fallback to default priority if model fails
15:30 — Logging, metrics, monitoring hooks

QA Report — MR !312
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Unit tests:      47/47 passed ✅
Integration:     12/12 passed ✅
E2E:             8/8 passed ✅
Load test:       p95 latency: 142ms ✅ (target: <200ms)
Model accuracy:  84.2% ✅ (target: >80%)
Coverage:        91% ✅
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Verdict: ALL CHECKS PASSED

Review — MR !312
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

1. [MEDIUM] predict_priority view missing rate limiting
   → Line 45: Add throttle_classes = [UserRateThrottle]

2. [LOW] Model loading happens on every request
   → Move to module-level singleton with lazy initialization

3. [LOW] Missing docstring on PriorityFeatureExtractor
   → Add class and method docstrings

Verdict: APPROVE with changes (no blockers)

# Merge to main triggers auto-deploy
# Vercel: frontend/preview → production promotion
# Render: backend auto-deploys from main branch

# Deploy verification
curl -s https://priority-predictor.onrender.com/api/health/
# {"status": "ok", "model_loaded": true, "version": "1.0.0", "db": "connected"}

# Create release with auto-generated notes
gh release create v1.0.0 \
  --title "Priority Predictor v1.0.0" \
  --generate-notes

# Changelog published, team notified on Discord + Telegram

GitLab (source) → CI/CD (lint + test) → Docker (sandboxed build)
     ↓
Vercel (preview) → QA (automated) → Code Review (Nico)
     ↓
Merge → Vercel (production) + Render (backend) → Health Check
     ↓
Release → Changelog → Discord + Telegram notification